基于KAZE特征匹配的指针式仪表自动读数方法

Auto-reading Method for Pointer Meter Based on KAZE Feature Matching

针对远距离多角度下的指针式仪表盘刻度模糊、表盘变形带来的自动识别困难,提出了一种基于KAZE特征的关键点匹配的表盘自动读数方法。首先针对每类表盘制作模板,然后分别提取待识别图像及其模板图像的KAZE特征,采用KNN进行关键点匹配,并通过RANSAC消除匹配错误的关键点。通过匹配结果得到表盘区域ROI,并根据ROI区域的直方图分布进行二值化和形态学处理得到边缘信息,采用概率霍夫变换拟合直线归类合并出指针直线,最终结合标注的先验知识得出读数。结果表明提出的方法很好的解决了远距离多角度下的表盘自动读数问题,在560个样本中,误差控制在一刻度之内的结果达到了96%以上,满足了实用要求。

Aiming at the problem of traditional auto-reading method for pointer meter that from long distance and large angle,an auto-reading method based on KAZE feature matching was proposed.Firstly,KAZE feature of the meter image and its template image were extracted,and then KNN feature matching and RANSAC which were used to reduce the bad matching points were applied to get the meter area（ ROI）.Histogram distribution of the ROI region for binarization and morphology for edge detection was gotten.Probability Hough transform was adopted to gain the lines,and the lines were combined to select the needle. Finally,the number of meter was calculated based on heuristic knowledge.Experimental results show that the proposed method can solve the problems of auto-reading method for pointer meter that from long distance and large angle,in the 560 samples,the error control within one scale reached more than 96%,satisfying the practical requirements.